Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75691.69161695443,
-      "required_pages_per_second_upper_estimate": 0.006560763990519082
+      "required_token_positions_per_second": 75644.55660022405,
+      "required_pages_per_second_upper_estimate": 0.006556678446731501
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9461.461452119303,
-      "required_pages_per_second_upper_estimate": 0.003774318434705323
+      "required_token_positions_per_second": 9455.569575028007,
+      "required_pages_per_second_upper_estimate": 0.003771968076842192
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18922.922904238607,
-      "required_pages_per_second_upper_estimate": 0.0043774851818269445
+      "required_token_positions_per_second": 18911.139150056013,
+      "required_pages_per_second_upper_estimate": 0.004374759217683792
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28384.38435635791,
-      "required_pages_per_second_upper_estimate": 0.0017284600859467276
+      "required_token_positions_per_second": 28366.70872508402,
+      "required_pages_per_second_upper_estimate": 0.001727383732739018
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28384.38435635791,
-      "required_pages_per_second_upper_estimate": 0.0021851084807572264
+      "required_token_positions_per_second": 28366.70872508402,
+      "required_pages_per_second_upper_estimate": 0.0021837477617324833
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28384.38435635791,
-      "required_pages_per_second_upper_estimate": 0.35243905728246533
+      "required_token_positions_per_second": 28366.70872508402,
+      "required_pages_per_second_upper_estimate": 0.35221958509857604
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08915428469535414,
-    "producer_p95_seconds": 0.2574945870001102,
-    "gpu_update_seconds": 0.9740567085368909,
-    "producer_batches_per_second": 11.216510831947826,
-    "gpu_batches_per_second": 1.026634272148362,
-    "producer_headroom": 10.925517622235484,
+    "producer_batch_seconds": 0.07069071313298991,
+    "producer_p95_seconds": 0.257033590998617,
+    "gpu_update_seconds": 0.9746636547775286,
+    "producer_batches_per_second": 14.146129748595229,
+    "gpu_batches_per_second": 1.025994962568143,
+    "producer_headroom": 13.787718521722947,
     "sustainable": true,
-    "buffer_seconds": 60.39151592928723,
+    "buffer_seconds": 60.42914659620677,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 427283.2303857373,
-    "upload_bytes_per_second": 4965065.723068627,
+    "generation_bytes_per_second": 429697.7371971901,
+    "upload_bytes_per_second": 4918652.573586773,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.68503459624375,
+    "estimated_full_shards_per_hour": 92.20313155113962,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.297586668999429,
-      "bytes": 7207958,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.0006021399985912,
       "bytes": 4975022,
@@ -562,6 +556,12 @@
       "bytes": 5636855,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.8648194890010927,
+      "bytes": 1161752,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -215,3 +215,4 @@
 {"step": 216500, "loss": 0.5662705302238464, "lr": 0.0003, "grad_norm": 0.16423504628066296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31688855500, "tokens_per_second": 149772.7716992007, "data_wait_s": 1.089499710360542e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 216600, "loss": 0.601808425039053, "lr": 0.0003, "grad_norm": 0.1728315363195944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31703583637, "tokens_per_second": 149834.48276005237, "data_wait_s": 1.4024997653905302e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 216700, "loss": 0.6148982718586922, "lr": 0.0003, "grad_norm": 0.15622664756895163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31718036413, "tokens_per_second": 146886.537276718, "data_wait_s": 1.09370012069121e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 216800, "loss": 0.5722600612789392, "lr": 0.0003, "grad_norm": 0.14606250506051432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31732506474, "tokens_per_second": 146420.55161636823, "data_wait_s": 1.002699718810618e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f6e06f7da492b08b1519cafca6ffda1633e06aa03493787a8a896cbcb4e55ea
-size 462903
+oid sha256:15f3d3d726a579853e2dd844478d81f5b3730c9fa781662d91a2f4bed8263d8f
+size 465021